Vapoursynth fmtconv

From wiki.selur.net
Jump to navigation Jump to search

fmtconv resizes in one pass with a classic interpolation kernel at high precision. Unlike the plain resize, the kernel is chosen on this page instead of on the Crop/Resize tab.

Filtering → Vapoursynth → Frame → Resize

fmtconv controls in Hybrid
fmtconv controls in Hybrid

What it does

fmtconv's resampler computes every output pixel from the surrounding source pixels, weighted by the chosen kernel. No neural network is involved, so it is fast and predictable; the kernel decides the balance of sharpness, ringing and aliasing.

Settings

Tick Resizer in the Resizer page and pick fmtconv as method.

Control Parameter Range / values Hybrid default Meaning
Resizer kernel point, rect, linear, cubic, lanczos, blackman, blackmanminlobe, spline36, spline64, spline144, spline, gauss, sinc spline64 Kernel that does the scaling. spline144 is written as the generic spline with 6 taps; the others use fmtconv's default taps (4) and kernel parameters.

Example

Default settings on a 10-bit YUV 4:2:0 source (640x356), resized to 1280x712:

# Resizing using fmtconv - spline64
clip = core.fmtc.resample(clip, kernel="spline64", w=1280, h=712, interlaced=False, interlacedd=False) # resolution 1280x712 before YUV420P10 after YUV420P16

How Hybrid calls it

  • fmtconv resizes to the target size of the Crop/Resize tab; separately filtered fields get half the target height, and interlaced content uses fmtconv's field handling according to the scan type.
  • 16-bit float input is converted to 32-bit float first; integer input comes out as 16-bit.
  • With an odd target height, limited-range YUV is converted to full range first.
  • Skipped when the size already matches (unless Lower resolution is on). In the crop view the Crop/Resize tab kernel is used instead.
  • Stepped Resize repeats the resize per step; Weighted also works with this method.

Background

fmtconv's resample

resample resizes and can change the chroma subsampling in the same step. Taps count half the number of lobes (1–128, default 4); the generic spline uses twice as many sample points as taps, so spline without extra taps gives the same result as spline64. cubic uses b = c = 1/3 by default. 8-bit input always gives 16-bit output.

Sources